Using a network of mobile sensors to track and map a dynamic spatio-temporal process in the environment is one of the current challenges in multi-agent systems. In this work, a distributed probabilistic multi-agent al...Using a network of mobile sensors to track and map a dynamic spatio-temporal process in the environment is one of the current challenges in multi-agent systems. In this work, a distributed probabilistic multi-agent algorithm inspired by the bacte- rium foraging behavior is presented. The novelty of the algorithm lies in being capable of tracking and mapping a spa- tio-temporal quantity without the need of machine learning, estimation algorithms or future planning. This is unlike most current techniques that rely heavily on machine learning to estimate the distribution as well as the profile of spatio-temporal quantities. The experimental studies carried out in this work show that the algorithm works well by following the concentration gradient of a dynamic plume created under diffusive conditions. Furthermore, the algorithm is inherently capable of finding the source of a diffusive spatio-temporal quantity as well as performing environmental exploration. It is computationally tractable for simple agents, shown to adapt to its environment and can deal successfully with noise in sensor readings as well as in robot dynamics.展开更多
文摘Using a network of mobile sensors to track and map a dynamic spatio-temporal process in the environment is one of the current challenges in multi-agent systems. In this work, a distributed probabilistic multi-agent algorithm inspired by the bacte- rium foraging behavior is presented. The novelty of the algorithm lies in being capable of tracking and mapping a spa- tio-temporal quantity without the need of machine learning, estimation algorithms or future planning. This is unlike most current techniques that rely heavily on machine learning to estimate the distribution as well as the profile of spatio-temporal quantities. The experimental studies carried out in this work show that the algorithm works well by following the concentration gradient of a dynamic plume created under diffusive conditions. Furthermore, the algorithm is inherently capable of finding the source of a diffusive spatio-temporal quantity as well as performing environmental exploration. It is computationally tractable for simple agents, shown to adapt to its environment and can deal successfully with noise in sensor readings as well as in robot dynamics.
